Low-light underwater image enhancement method based on brightness estimation

Active illumination in underwater imaging encounters two significant degradation issues: color distortion and uneven lighting, which complicate the analysis of underwater images and videos. This study introduces a network for enhancing underwater low-light images that utilizes techniques for color correction and brightness estimation. The network comprises four sub-networks: the Image Decomposition Network, which utilizes an encoder-decoder architecture with dilated convolutions for global luminance estimation, generating illumination and reflection images; the Illumination-Guided Enhancement Block, which performs channel-wise concatenation of the illumination and reflection images; the Noise Suppression and Fusion Model, which reduces noise in the image using the illumination map; and the Color Restoration Network, which handles color correction. Results from experiments conducted on real-world data show that the proposed method effectively mitigates issues related to uneven lighting and color deviation in underwater low-light images, surpassing conventional methods in both objective evaluation metrics and subjective visual quality.